Session Item

Diffusion-weighted imaging (DWI) is being explored in MR-LINAC (MRL) empowered MR-guided radiotherapy in prostate cancer (PC), while its positional and geometric reliability might be a concern due to the distortion of DWI images. In this study, we aim to quantitatively evaluate the positional and geometric reliability of DWI by referring to the anatomical T2W-MRI obtained in the same scan on a 1.5T MRL and a 1.5T MR-simulator (MRsim).

8 patients with localized PC were included. T2W (voxel = 1.2x1.2x2mm³) and DWI (voxel = 2.5x2.5x4mm³) planning MR images acquired on a 1.5T MRsim and their corresponding daily MRI images at the first fraction on a 1.5T MRL (same sequence and voxel ). The prostate was manually delineated. The position and shape deviation between T2W and DWI images were evaluated using centroid shift (∆T2w-DWI, |T2W centroid position - DWI centroid position|), volume and dice similarity coefficient (DSC) on MRsim and MRL images, and compared using the non-parametric signed-rank test. The apparent diffusion coefficient (ADC) values were also calculated and compared between MRsim and MRL.

Sub-voxel deviation of ∆T2w-DWI along LR (MRsim: 0.45±1.18mm vs MRL: 0.52±1.43mm), AP (0.63±1.28 vs 1.37±1.68) and SI (1.06±1.68mm vs 1.90±1.97mm) directions were observed between MRsim and MRL (all p>0.05). Insignificantly different DSC was also obtained between T2W and DWI for MRsim (0.82±0.03) and MRL (0.82±0.03, p>0.05).  Although smaller prostate volume was noted in T2W than in DWI for MRsim (39.22±19.65mm³ vs 42.23±19.65 mm³  p<0.05) and MRL (40.05±19.63mm³ vs 41.84±17.60mm³, p=0.328), such difference might be negligible, which was probably due to the larger DWI voxel . Meanwhile, an insignificantly larger ADC was observed in MRsim DWI (1.1±0.1 mm³/ms) than MRL DWI (1.0±0.8 mm³/ms, p>0.05), despite the compromised gradient coil configuration on the MRL. The result of this study is mainly limited by the small sample and the lack of ground truth of geometry. Future studies with increased sample and phantom validation are warranted to consolidate the findings in the current study.

Table 1: The dice coefficient, centroid positional shift and prostate volume delineated using T2W and DWI images were illustrated. The MRL and MRsim T2W were also linearly registered to estimate the reference DSC and centroid positional shift.

Figure 1: Similar prostate contours were obtained using all 4 scans.

In our preliminary results, the volumetric and positional deviations in DWI from the reference T2W images on both MRsim and MRL were small, suggesting that the use of geometrical distortion in DWI images might not much compromise the geometric accuracy in prostate MRgRT. Meanwhile, DWI ADC was consistent between MRsim and MRL, so it might serve as a potential quantitative imaging biomarker in MRgRT.